JadidJust-in AI Developments, Incidents & Deadlines
May 2026
SMTWTFS12345678910111213141516171819202122232425262728293031

AI news, Friday, May 15, 2026

12 stories

TLDR

  • U.S. clears Nvidia H200 chip sales to 10 Chinese firms (May 14) — After Jensen Huang joined Trump's China summit as a last-minute addition, the Commerce Dept approved H200 exports to Alibaba, Tencent, ByteDance, JD.com, Lenovo, and five others (up to 75K chips each). Actual deliveries remain stalled as Chinese firms await Beijing guidance.
  • Cisco cuts ~4,000 jobs (5%) and posts record $15.8B quarterly revenue (May 14) — Restructuring frees cash for silicon, optics, and security AI; the company now projects $9B in AI orders for FY2026 (up from earlier $5B estimate). Stock surged 17% after-hours.
  • U.S. and China to launch formal AI safety talks (May 14) — Treasury Secretary Bessent announced a bilateral AI safety protocol after Trump–Xi meetings; Jensen Huang's H200 deal and a fresh AI-safety framework from Washington's two AI superpowers in one day is the biggest geopolitical AI story of the week.
  • xAI retires multiple Grok models TODAY at noon PT (May 15) — grok-4-0709, grok-3, grok-code-fast-1, grok-imagine-image-pro, and others go dark; slugs silently redirect to grok-4.3 at different pricing. API users need to audit call volumes now.
  • OpenAI launches $4B Deployment Company, acquires Tomoro (May 12) — A majority-owned subsidiary backed by TPG, Bain Capital, Goldman Sachs and 16 others will embed Forward Deployed Engineers directly inside enterprise clients; concurrent Tomoro acquisition brings 150 FDEs on day one.

Action Items

Things your engineering team should look at or do this week.

  • Audit xAI API usage before noon PT today — grok-4-0709, grok-4-fast, grok-4-1-fast, grok-3, grok-code-fast-1, and grok-imagine-image-pro retire at 12:00 PM PT May 15; traffic automatically falls back to grok-4.3 at a different price point. Check call volumes and update model strings to explicit IDs. (xAI Migration Docs | Model card)
  • Evaluate Claude for Small Business for any SMB-facing product work — Anthropic's new package ships 15 prebuilt agenticDescribes AI systems where a model uses tools and decides its own steps instead of following a fixed script.Anthropic ↗ workflows (payroll, invoicing, contract review, campaign creation) plus native connectors for QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace, and M365 — worth checking before building any of these integrations from scratch. (TechCrunch | Axios)
  • Watch the H200 China supply chain story closely — The U.S. cleared H200 exports to 10 Chinese buyers (75K chip cap each), but zero shipments have moved yet as Beijing reviews orders. If your 2026 GPUGraphics processing unitA highly parallel processor that does the large matrix math behind training and running models.NVIDIA ↗ capacity planning relies on Nvidia supply remaining tight, this is a meaningful demand-side variable. (CNBC)

Models & Tools

  • Anthropic Claude for Small Business (May 13) [Agents] [Customer AI] — Anthropic launched a bundle of 15 ready-to-run agentic workflows and 8 native app connectors (QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace, M365, and more) targeting the 44%-of-GDP small business segment that has lagged enterprise AI adoption; a 10-city roadshow beginning in Chicago kicks off this month. (TechCrunch | SiliconANGLE)

  • OpenAI Deployment Company + Tomoro acquisition (May 12) [Agents] [Customer AI] — OpenAI's new majority-owned subsidiary pairs $4B from 19 PE and consulting partners (TPG, Bain Capital, Goldman Sachs, McKinsey, Capgemini) with Forward Deployed Engineers who embed directly in client operations; the concurrent Tomoro acquisition seeds 150 FDEs on day one. This is OpenAI's most direct move yet into the systems-integrator space — watch for competitive pressure on Accenture/Deloitte AI practices. (OpenAI | Axios)

  • NVIDIA + IREN: 5 GW AI infrastructure partnership (May 8) [Infra] — NVIDIA backed AI cloud operator IREN to deploy up to 5 GW of AI infrastructure globally, with early focus on IREN's 2 GW Sweetwater campus in Texas; the deal is multigenerational and covers chips, networking, and software. Adds to the growing pattern of NVIDIA taking equity-adjacent stakes in neocloud operators to secure GPU deployment capacity. (Digitimes)

APIs & Pricing

  • xAI: 8 Grok models retire today at noon PT (May 15) [Pricing] — Retired slugs: grok-4-1-fast-reasoning, grok-4-1-fast-non-reasoning, grok-4-fast-reasoning, grok-4-fast-non-reasoning, grok-4-0709, grok-3, grok-code-fast-1, grok-imagine-image-pro. Slugs remain active as aliases to grok-4.3 ($1.25/$2.50 per 1M tokensA small unit of data, such as part of a word, that AI models read and generate. Many AI services price by tokens in and out.NVIDIA ↗) — so no breakage, but surprise billing if your existing models were cheaper. Migration is a single model-string change. (xAI Docs)

Security

  • arXiv enforces 1-year ban for unchecked AI-generated errors in submissions (May 15) — A clarification from arXiv moderator Thomas Dietterich confirmed submissions with "incontrovertible evidence" of unreviewed LLM-generated content now face a one-year ban plus a requirement that future submissions must first be accepted by peer-reviewed venues. Combined with the earlier ban on unreviewed CS review/position papers, arXiv is now the first major academic platform to impose hard penalties — relevant if your team publishes research or cites preprints. (Startup Fortune | 404 Media)

Industry & Business

  • U.S. clears Nvidia H200 chip sales to 10 Chinese firms; Jensen Huang joins Trump's China summit (May 14) [Infra] [Trending] [Policy] — The Commerce Department approved H200 exports (up to 75K chips each) to Alibaba, Tencent, ByteDance, JD.com, Lenovo, and five others on the same day Jensen Huang flew to Alaska to board Air Force One after a personal call from Trump. Despite the approval, Chinese firms are awaiting Beijing guidance and no shipments have moved. Nvidia shares gained on the news; the $50B/year China AI chip market now has a structural opening — though it remains politically fragile. (CNBC H200 | CNBC Huang trip | Fast Company)

  • Cisco cuts ~4,000 jobs (5% of workforce) while posting record revenue (May 14) [Hiring] [Trending] — Layoff notifications began May 14 against the backdrop of record Q3 FY2026 revenue of $15.8B and $5.3B in AI infrastructure orders; the company raised its FY2026 AI-order forecast to ~$9B. Investment is being redirected to silicon/optics, security, and AI tooling. This is the Cloudflare pattern one week later: record revenue, AI-driven role-mix change, stock up. (TechCrunch | Tech Startups)

  • LinkedIn cuts 875 jobs (5%) in AI-driven restructuring (May 13) [Hiring] — CEO Daniel Shapero announced cuts across Global Business Organisation, marketing, engineering, and product teams, explicitly framing the move as building "smaller, faster, more agile teams that heavily leverage AI tools" — not direct AI job displacement. Notable because LinkedIn revenue grew 12% YoY; adds to the 2026 tech-layoff wave now past 100K workers. (Republic World | GeekWire)

  • Wirestock raises $23M to supply multimodal training data to AI labs (May 14) [Data] — Pivoting from stock-asset marketplace to structured AI training data supplier, Wirestock now works with six of the largest foundation model makers and has onboarded 700K+ artists and designers; $40M ARR and growing. A useful reminder that proprietary, high-quality multimodal data is increasingly a competitive moat. (Tech Startups)

AI in Production

  • White House AI executive order stalled by internal infighting (May 13) [Policy] [Agents] — Two weeks after Anthropic's Claude Mythos threw Washington into a reactive posture, no federal AI regulation has emerged: Trump administration officials remain publicly contradictory on the scope of pre-release model vetting, and Trump's China summit has consumed White House bandwidth. CAISI (the federal AI testing body) has ~30 staff and limited authority. Operational implication: the U.S. regulatory environment for frontier models remains unresolved for at least another few weeks. (Axios | Federal News Network)

Policy & Safety

  • U.S. and China announce bilateral AI safety protocol after Trump–Xi summit (May 14) [Trending] — Treasury Secretary Bessent confirmed Washington and Beijing will begin formal discussions on AI best practices and safeguards, explicitly targeting "non-state actor" access to frontier models. Bessent: "The two AI superpowers are gonna start talking." This is the first bilateral AI safety framework discussed at head-of-state level since the 2023 Bletchley Declaration; momentum depends entirely on broader trade-deal durability. (CNBC | Epoch Times)

Paper Spotlight

  • RAVEN: Real-time video generation via causal autoregressive extrapolation (May 14) — MVP Lab's RAVEN achieves real-time video generation using causal autoregressive extrapolation with improved training alignment for temporal consistency. Pairs with CM-GRPO (consistency model + GRPO reinforcement learningReinforcement learningTraining by trial and error, where the model takes actions and adjusts based on the rewards it gets.Kaelbling et al. 1996 ↗) for higher-fidelity sampling. Relevant for teams building video-generation pipelines or evaluating alternatives to diffusion-based approaches. (arXiv)

Type to search · ↑↓ to move · enter to open